Table 1.

Performance of GBDT models trained using different features on the DUD-E test set (N = 12).

Features used in the model	Average number of active compound in top N, with 90% confidence interval			Diversity*b	AUROC*d
	N = 10	N = 50	N = 100
GlideSP and multi-resolution ExptGMS*a	5.4 [5.0, 5.8]	21.6 [20.4, 22.9]	33.2 [30.9, 35.1]	0.64 [0.60, 0.67]	0.66
GlideSP and TF3P	5.4 [4.9, 5.8]	19.5 [18.1, 20.8]	27.7 [26.0, 29.6]	0.62 [0.60, 0.64]	0.64
GlideSP and MM/GBSA	4.5 [4.0, 5.2]	18.5 [17.0, 19.9]	30.1 [28.1, 32.2]	0.66 [0.62, 0.68]	0.65
GlideSP and USRCAT	4.5 [3.9, 5.1]	18.1 [16.9, 19.3]	30.4 [28.0, 32.3]	0.63 [0.59, 0.66]	0.63
GlideSP and Alpha sphere	5.1 [4.6,5.6]	18.4 [17.0, 19.8]	29.2 [27.2, 31.0]	0.64 [0.61, 0.68]	0.63
GlideSP*c	4.3 [3.8, 4.8]	17.3 [16.1, 18.6]	29.9 [27.9, 31.9]	0.66 [0.62, 0.68]	0.62

^*aMultiresolution indicates ExptGMS at resolutions of 2.5, 3.0, 3.5, 4.5, and 5.5 Å.

^*bDiversity of active compounds among the top 100 ranked compounds.

^*cRanking was performed directly using the GlideSP score without using a trained model.

^*dAUROC was calculated by using GBDT predicted probability of being an active compound.